Torsten Sowa

An AC power flow linearization for power system optimization using linear programming

This paper presents a method for the linearization of the non-linear power flow equations, which can be used in mixed integer linear optimizations. The power flow equations are linearized around an operating point using the Taylor approximation. The linearization implies an approximation error, which can be reduced iteratively by modifying the operating point. In addition to existing approaches, controllable assets like voltage regulated transformers or phase shifters are integrated into the linearized grid constraints. The model is exemplarily applied to an operation planning model of distributed energy resources considering grid restrictions. The results show that this approach reduces the approximation error significantly and that it is robust for different real distribution grids as well as for different generation and load scenarios.

Method for modelling fluctuating energy generation plants for the assessment of their impact on medium and low voltage grids

This paper describes a method for taking the forecasting uncertainty into account when assessing the impact of volatile generation on power grids. To this end the generation of feed-in scenarios for generation plants that include the uncertainty of the weather forecast is described. With a three-step model that firstly forecasts local weather parameters and, secondly, generates scenarios for these parameters, thirdly, a reduced number of resulting feed-in profile scenarios for each plant is computed. With these scenarios the assessment of the grid-impact of the plants can be calculated taking into account the uncertainty of the prognosis. In a case study, feed-in profiles for the plants in an exemplary region are generated which can be used for an assessment of the grid impact in this region using probabilistic load flow calculation.

Modelling the charging probability of electric vehicles as a gaussian mixture model for a convolution based power flow analysis

This paper presents an approach for modelling the charging probability of electric vehicles as a Gaussian mixture model. The model is built up by assembling adapted multivariate normal probability density functions. This is done because the expectation maximization algorithm fails finding maximum likelihood estimates in respect of the charging power of the generated charging profiles. This Gaussian mixture model enables for capturing the charging profiles comprehensively with a few parameters and therefore it enables for calculating the charging probability dynamically for individual parameter intervals. The underlying assumptions about battery capacity, consumption, charging infrastructure, type of weekday and settlement structure determine the generation of the charging profiles. The proposed approach makes these parameters available for the density. Thereby, the provision of the charging profiles gets obsolete. This density can be used for a convolution based power flow analysis which offers benefits regarding the computational effort and random access memory usage compared to Monte Carlo-like simulations.